core.trainers.grpo.trainer

core.trainers.grpo.trainer

Axolotl GRPO trainers (with and without sequence parallelism handling)

Classes

Name Description
AxolotlGRPOSequenceParallelTrainer Extend the base GRPOTrainer for sequence parallelism handling
AxolotlGRPOTrainer Extend the base GRPOTrainer for axolotl helpers

AxolotlGRPOSequenceParallelTrainer

core.trainers.grpo.trainer.AxolotlGRPOSequenceParallelTrainer(
    model,
    reward_funcs,
    args=None,
    train_dataset=None,
    eval_dataset=None,
    processing_class=None,
    reward_processing_classes=None,
    callbacks=None,
    optimizers=(None, None),
    peft_config=None,
    optimizer_cls_and_kwargs=None,
)

Extend the base GRPOTrainer for sequence parallelism handling

Methods

Name Description
get_train_dataloader Get dataloader for training
get_train_dataloader
core.trainers.grpo.trainer.AxolotlGRPOSequenceParallelTrainer.get_train_dataloader(
)

Get dataloader for training

AxolotlGRPOTrainer

core.trainers.grpo.trainer.AxolotlGRPOTrainer(*args, **kwargs)

Extend the base GRPOTrainer for axolotl helpers